Public infrastructure · computational social science · 2016–2025

Philippine DPWH Flood-Control Network Science

Two complementary studies of procurement structure and project-similarity relationships, built from reproducible DPWH Transparency data and presented through precomputed network evidence.

Explore contractor network Explore project relationships

Overview

One dataset, two research questions

Metrics below are calculated in the browser from the generated V2 CSV outputs—never inserted as invented constants.

Loading overview datasets…
Projects—Unique contract IDs
Contractors—Canonical PCAB identities
Marked revoked—Source marker, not current status
DPWH offices—Normalized implementing offices
Years covered—Infrastructure years
Valid coordinates—Non-zero, global-range pairs
Project 1

Contractor ↔ DPWH Office Network

How is flood-control procurement structured across contractors and implementing offices, where is activity concentrated, and how has the network evolved from 2016–2025?

Open Project 1
Project 2

Project ↔ Project Potential Overlap Network

Which flood-control projects are geographically, textually, and temporally related enough to warrant detailed project-level comparison?

Open Project 2

Project 1 · Procurement structure

Contractor ↔ DPWH Office Network

A weighted temporal bipartite network: a relationship exists when a contractor won at least one flood-control project implemented by a DPWH office.

Contractors ↔ Offices
Project 1 loads when this section enters view.

Full network

Interactive procurement network

Contractor DPWH office
Visible nodes— Visible edges— Contracts— Contract value—

Zoom with the wheel, drag to pan, and click a node for metrics. Labels prioritize hovered, selected, and structurally important nodes.

Temporal playback

Procurement relationships through time

2016 2025 All years

Full network vs filtered network

Overall ecosystem and high-activity core

The core uses the data-adaptive activity rule documented by Notebook 02.

Loading comparison metrics…

Top hubs

Structurally important contractors

Temporal network

Annual evolution, 2016–2025

Robustness tests

How quickly does connectivity fragment?

Random contractor removal is compared with targeted removal by structural importance. The curve reports the largest connected-component proportion.

Full network

Overall procurement structure

Shows the complete observed contractor–office ecosystem and the breadth of implementing relationships.

High-activity core

Sustained and high-volume activity

Focuses attention on contractors meeting several observed activity thresholds without implying wrongdoing.

Projection and communities

Shared-office similarity

Connects contractors that work with overlapping offices. Communities are structural groups, not collusive groups.

Project 2 · Spatial, textual, and temporal relationships

Project ↔ Project Potential Overlap Network

A project-level candidate network for prioritizing detailed comparison. Relationships preserve their individual evidence and are not confirmed duplicates.

Projects ↔ Projects
Project 2 loads when this section enters view.

All projects and review statuses

Interactive project-similarity network

Visible projects— Visible relationships— Exact-coordinate links— Temporal-overlap links—

Click a node for project details or an edge for transparent pair evidence and comparison.

Full network vs filtered network

Broad candidates and independent evidence

The high-evidence rule combines spatial, textual, and temporal indicators derived in Notebook 03.

Loading comparison metrics…

Most connected project candidates

Candidate-network hubs

Component explorer

High-evidence project clusters

Full candidate network

Broad comparison universe

Captures nearby projects that are plausible candidates for further text and timing comparison.

High-evidence subgraph

Independent indicators

Retains relationships supported by spatial, textual, and temporal evidence. It does not confirm duplication.

Connected components

Project clusters

Groups projects directly or indirectly related through high-evidence candidate relationships.

Project map

Coordinate coverage and reuse

Exact coordinate matches are aggregated before marker clustering so reused locations remain inspectable.

Map loads when this section enters view.

Methodology

From raw DPWH records to network evidence

01Raw DPWH JSONLocal authoritative source
02Canonical cleaningProjects, PCAB IDs, offices, dates, values, coordinates
03Two network modelsContractor–office and project similarity
04Static browser layerLoad, filter, visualize, compare, explain

Project 1 definition

Weighted bipartite edges represent won projects between canonical contractor PCAB identities and normalized implementing offices. Value and contract counts are retained separately.

Project 2 definition

BallTree candidate generation limits comparison to nearby coordinates. TF-IDF text similarity and implementation timing then provide independent pair evidence.

Identity and revoked marker

PCAB IDs and internal node IDs drive joins. The “(revoked)” suffix is display-only; “not marked revoked” does not assert current active status.

Limitations

What these networks cannot determine

Source completeness

Results inherit missing, malformed, delayed, or inconsistent values in the public source records.

Coordinates

Exact coordinate reuse may represent shared sites, coarse geocoding, linear infrastructure, or source inconsistency. It is not automatic duplication.

Network measures

Centrality, concentration, communities, and robustness describe structure—not intent, legality, or misconduct.

Review language

“High-priority review” means several indicators justify detailed comparison. It is not a probability of fraud.

The analyses identify patterns that may warrant examination. They do not establish wrongdoing.